On-device AI video restoration for macOS
Lumasce repairs and upgrades your video on your own Mac. Remove watermarks, logos and burned-in subtitles, restore mosaic or pixelated footage, upscale to 2×/4×, generate and translate subtitles, and interpolate frames — without uploading a single frame.
What it does
Restoration & inpainting
Erase watermarks, logos and burned-in subtitles by inpainting (reconstructing the pixels behind them) instead of cropping the frame.
Mosaic & pixelation repair
Mosaic is downsampling: outlines, low-frequency structure and inter-frame differences survive. Lumasce uses those residuals to reconstruct detail — light to moderate mosaics often come back close to the original.
Enhancement
Upscale, denoise, restore faded color and interpolate frames for smoother motion.
Subtitles
Generate subtitles from speech, review, then translate into bilingual output (original on top, translation below).
Smart mode
Chain several steps into one run and preview the plan before you commit.
Private by design
Processing happens locally. The cloud only handles your account, licence and billing — never your footage.
How it works
- Pick your clips. Drop in one file or a whole folder.
- Choose a job. Restoration, mosaic repair, enhancement or subtitles — or chain them in Smart mode.
- Run it locally. Models run on your Mac's GPU; nothing is uploaded.
- Compare and export. Check before/after and keep the original or the repaired version.
Why on-device instead of cloud?
| On-device (Lumasce) | Cloud service | |
|---|---|---|
| Privacy | Footage never leaves your Mac | You upload the original |
| Cost | One-time licence, no per-minute fees | Billed per minute of compute |
| Speed | Depends on your chip | Depends on your connection and their queue |
| Best for | Sensitive or large libraries, long-term use | Occasional jobs on a weak machine |
What it can and cannot do
We would rather set expectations than oversell:
- Light to moderate mosaic: reconstruction. Because mosaic is downsampling, plenty of structure survives across frames, so results often come back close to the original picture.
- Heavy mosaic or wide occlusion: inference. When too little survives, the model produces the most plausible image — natural-looking, but not guaranteed to match the original pixel for pixel.
- Watermark removal depends on the background. Flat areas come back clean; busy textures such as faces or dense text may show traces, in which case cropping is the safer call.
- Never re-process the same file repeatedly — each pass adds its own reconstruction and re-encoding loss.
Frequently asked questions
Is my video uploaded anywhere?
No. Video processing runs locally on your Mac. The cloud is used only for your account, licence and billing; your footage is never uploaded.
Can mosaic removal recover the original picture?
It depends on how much information survived. Mosaic is downsampling, so outlines, low-frequency structure and inter-frame differences often remain — for light to moderate mosaics the result can come close to the original. For heavy mosaic or wide occlusion there is too little left, so the model can only infer a plausible image.
Which Macs are supported?
Apple silicon Macs (M-series) running macOS. A Windows build is planned.
Do I need an internet connection?
You need to sign in and activate once. After that, processing runs locally.
Can it remove subtitles burned into the picture?
Yes — burned-in subtitles are treated like a watermark and removed by inpainting, which keeps the full frame instead of cropping it away.
What output formats are supported?
Common formats in and out, including H.264 and HEVC, with adjustable quality settings.
Read the guides for the technical detail behind these claims, or get in touch.